Attestation of origin controller for machine learning models

The framework addresses the challenge of verifying ML model trustworthiness in MLaaS environments by employing watermarking and attestation protocols, ensuring secure and legitimate use of ML models through dynamic verification and ownership validation.

EP4764921A1Pending Publication Date: 2026-06-24THALES DIS FRANCE SA +1
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
EP · EP
Patent Type
Applications
Current Assignee / Owner
THALES DIS FRANCE SA
Filing Date
2024-12-20
Publication Date
2026-06-24

AI Technical Summary

Technical Problem

Existing technologies lack clear methods for verifying the trustworthiness and ownership of machine learning models in cloud applications, particularly in Machine Learning as a Service (MLaaS) environments, where models can be maliciously altered or tampered with, posing risks to system functionality and security.

Method used

A framework and protocol for dynamic verification of ML model trustworthiness in a zero-trust and distributed environment, utilizing ML model watermarking and attestation of origin (AO) to ensure the legitimacy and ownership of ML models, involving an attestation protocol that includes an Attester Agent, ID Agent, Verifier, and Challenger to validate watermark labels and provenance.

Benefits of technology

Provides secure, real-time verification of ML model ownership and trustworthiness, ensuring that cloud applications can confidently utilize legitimate models by validating their origin and integrity, thereby preventing misuse and tampering.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure IMGAF001_ABST
    Figure IMGAF001_ABST
Patent Text Reader

Abstract

Provided is a system and method, implemented by way of a framework and protocol, for dynamic verification of an ML model trustworthiness in a zero-trust and distributed environment, based on ML model watermarking in conjunction with the production of an attestation. The attestation protocol for a MLaaS function verifies the origin and ownership of a trained ML model used therein. An attestation of origin (AO) result is produced to provide ML model ownership information, whereby a cloud application can evaluate the AO result to verify the ML model ownership information as to whether it can trust the ML model, for example, according to its policy and decide whether to use it or not. Other embodiments disclosed.
Need to check novelty before this filing date? Find Prior Art